JP2012160063A - Sphere detection method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a sphere detection method capable of detecting a sphere at a much higher speed on the basis of shape data in the case of measuring a sphere as a three-dimensional point group.SOLUTION: The sphere detection method includes: a shape data acquisition step (S1) of acquiring shape data on a sphere as a measurement point group by three-dimension measurement means; a normal vector calculation step (S2) of calculating a normal vector of each measurement point in a virtual coordinate space; a measurement point movement step (S32) of moving the measurement points for the radius of the sphere to a direction opposite to the normal vector; a projection step (S34) of projecting the moved measurement points to a projection surface which is vertical to an optical axis of the three-dimension measurement means; a dense area search step (S44) of searching a dense area where the projected moved measurement points are dense from the projection surface; and a spherical center estimation step (S45) of, from the moved measurement point group in the virtual coordinate space corresponding to the searched dense area, estimating the center of the sphere in the virtual coordinate space.

Description

  The present invention relates to a sphere detection method, and more particularly, to a sphere detection method for detecting a sphere using virtual space coordinates based on sphere shape data obtained by measuring or imaging a sphere.

Conventionally, as a method of detecting a target object, there is a method of acquiring shape data of the target object by measuring the distance from the target object to imaging or a measurement point to the target object. Based on the shape data, the object area occupied by the object in the two-dimensional or three-dimensional virtual space coordinates is generated, thereby expressing the shape of the object in the virtual coordinate space.
However, in the two-dimensional or three-dimensional virtual coordinate space, there is a problem that an unclear shape portion is generated in the vicinity of the boundary of the object region generated as described above due to noise or the like during measurement or imaging.
Therefore, a method is known in which a sphere is sequentially applied to an object boundary pixel in a three-dimensional image, and whether or not the pixel of interest is extracted is determined based on the overlapping ratio of the object and the sphere (see Patent Document 1). ).

Japanese Patent No. 3888975

By the way, even when the object is measured as a three-dimensional point group, it is necessary to correct the vicinity of the boundary of the sphere in the measured shape data. This is the same when the object is a sphere.
In this regard, the conventional technique disclosed in Patent Document 1 performs pixel extraction processing on voxel data that constitutes a three-dimensional image of a computed tomography apparatus (CT apparatus). When measuring as a point cloud, there is a problem that the conventional technique cannot be applied to the point cloud.

Further, as another correction method, a method using a least square method or a Hough transform is known, but in the least square method, when an outlier such as noise exists at measurement points constituting a three-dimensional point group, Since the detection accuracy of a sphere falls, it is not preferable.
Further, when the Hough transform is used, a large amount of memory is required to store the feature points, and since a large amount of calculation requires a long time for detecting a sphere, it is not preferable.
The present invention has been made to solve the above-described problems, and an object of the present invention is to detect a sphere at a higher speed based on shape data when the sphere is measured as a three-dimensional point cloud. It is an object of the present invention to provide a method for detecting a sphere that is possible.

  In order to achieve the above object, a method for detecting a sphere according to claim 1 is a method for detecting a sphere that represents a shape of a sphere as a target object in a virtual coordinate space, and the sphere as a measurement point group by a three-dimensional measuring unit. A shape data acquisition step of acquiring the shape data of the measurement data, a normal vector calculation step of calculating a normal vector of each measurement point of the measurement point group constituting the shape data in the virtual coordinate space, and the normal vector, In the opposite direction, the measurement point moving step for moving each measurement point by the radius of the sphere, and the moved measurement point moved by the measurement point moving step on the projection plane perpendicular to the optical axis of the three-dimensional measurement means A projection step for projecting; a dense region search step for searching from the projection surface for a dense region where the measurement points after movement projected by the projection step are dense; and the dense region search step. A spherical center estimation step of estimating the center of the sphere in the virtual coordinate space from the measurement point group after the movement in the virtual coordinate space before the projection corresponding to the dense area searched from the step. And

The sphere detection method according to claim 2, wherein the dense region search step generates a projection region in which the projection plane is partitioned in a lattice shape, and searches the dense region from each projection region. It is characterized by.
The sphere detection method according to claim 3, wherein in the sphere center estimation step, the estimation is performed by obtaining an average of coordinate values of the measurement point group after the movement in the dense area. And

  According to a sphere detection method of claim 4, in the sphere center estimation step according to claim 1 or 2, the sphere center estimation step performs estimation by a least square method on the moved measurement point group included in the dense area of a virtual coordinate space. It includes a step of improving accuracy.

According to the sphere detection method of the present invention, in the virtual coordinate space, a normal vector calculation step of calculating a normal vector of each measurement point constituting the shape data composed of the measurement point group acquired by the three-dimensional measurement means; A measuring point moving step for moving the measuring point by the radius of the sphere in the direction opposite to the normal vector, a projecting step for projecting the moved measuring point onto a projection plane perpendicular to the optical axis of the three-dimensional measuring means, and a projection A dense area search step for searching for a dense area where the measured points after movement are dense, and a sphere for estimating the center of the sphere from the measured points in the virtual coordinate space before projection corresponding to the searched dense area A center estimation step.
As a result, the amount of calculation required for estimating the center of the sphere is proportional to the number of measurement points, and the amount of calculation can be further reduced, so that the estimation can be performed at a higher speed.

It is a schematic block diagram of the target object recognition apparatus used for embodiment of this invention. It is a flowchart of the target object detection method which concerns on this invention. (A) is a diagram showing a measurement point group acquired by a three-dimensional measurement means, (B) is a diagram extracting any four adjacent points from the measurement point group shown in (A), and (C) is a diagram of (B). It is a figure which shows the triangular mesh produced | generated from four points. It is the schematic of the projection process shown in FIG. It is a flowchart which shows the projection processing routine performed with the target object detection method shown in FIG. (A) is a diagram showing a schematic diagram of projection from the three-dimensional virtual coordinate space onto the projection plane in the projection processing shown in FIG. 2, and (B) is a measurement point after movement projected onto the projection plane in (A). It is a figure which shows a group. 3 is a flowchart illustrating a spherical center estimation routine performed by the object detection method illustrated in FIG. 2. (A) is the schematic which divides the projection surface of FIG. 6 (B) in a grid | lattice form, (B) is the schematic which shows estimation of the spherical center in virtual coordinate space from the measurement point after the movement in a projection surface.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a schematic configuration diagram of an object recognition device 1 used in an embodiment of the present invention.
As shown in FIG. 1, the object recognition apparatus 1 is an apparatus that expresses the shape of a sphere 2 that is an object in a virtual coordinate space, and obtains shape data of the sphere 2 supported by a support 3. Means 4, calculation means 6 for calculating the sphere 2 in the virtual coordinate space from the shape data acquired by the three-dimensional measurement means 4, a memory device 8 for storing the results calculated by the calculation means 6, and the like External output means 10 for outputting the result.

Here, the three-dimensional measuring means 4 acquires the shape data of the sphere 2 by measuring the distance to the measurement point on the sphere 2.
The computing means 6 and the memory device 8 are provided in a computer (not shown). The computing means 6 is constituted by a central processing unit (hereinafter abbreviated as CPU) of the computer, and the memory device 8 is constituted by including RAM, ROM and the like.
The external output means 10 outputs a result calculated by serial communication, LAN network communication, screen output, external memory writing, file output, or the like.

Hereinafter, the sphere detection method of the target object recognition apparatus 1 according to the present invention configured as described above will be described.
FIG. 2 shows a flowchart of the object detection method. Each process after step S2 described below is performed by executing each program stored in the memory device 8 by the calculation means 6. It is assumed that the radius of the sphere 2 is stored in the memory device 8 in advance.

In step S1, the shape data of the sphere 2 is acquired as a measurement point group from the three-dimensional measurement means 4 (shape data acquisition step). The three-dimensional measuring means 4 uses, for example, a laser distance meter, a stereo camera, a ToF (Time-of-Flight) system camera, or the like.
When the three-dimensional measuring unit 4 is a laser distance meter, the distance from the position of the three-dimensional measuring unit 4 to each measurement point on the sphere 2 is measured by laser emission and measurement of the reflected laser. Shape data is acquired using each distance data as a three-dimensional measurement point group.
Further, when the three-dimensional measuring unit 4 is a ToF camera, the sphere 2 is acquired as a three-dimensional distance image by photographing the sphere 2.
When the three-dimensional measuring unit 4 is a stereo camera, the sphere 2 included in the image data is acquired as a three-dimensional measurement point group by capturing the sphere 2. Here, it is difficult for a stereo camera to measure an object having no pattern, but data can be acquired by combining projection of pattern light.

In step S2, a normal vector of each measurement point in the virtual coordinate space acquired in step S1 is calculated (normal vector calculation step).
Specifically, FIG. 3 shows a process of generating a triangular mesh from the measurement point group acquired by the three-dimensional measuring means 4, and will be described below with reference to FIG.

For example, a measurement point _{P 1} (A), the select and measurement point _P 2 to P ₄ adjacent to the measurement point _{P 1,} shown in (C) from the selected four points as (B) Such a triangular mesh is generated.
After generating the triangular mesh for all measurement points, the normal vector of each measurement point is calculated. Here, assuming that a set of triangles having the measurement point P _i as a vertex is F and a normal vector of the triangle f is n _f , the normal vector n _pi of the vertex P _i is calculated from the following equation (1).

The function N (a) represents unit vectorization of a.
Further, when noise is included in the measurement data, since the normal vector includes an error, smoothing of the normal vector calculated from the above equation 1 is performed. Here, assuming that a set of triangles having the measurement point P _i as a vertex is F and a normal vector of the vertex k of the triangle f is n _{f_k} , the normal vector n _Pi of the smoothed vertex P _i is expressed by the following equation 2. Calculated from the formula.

The noise contained in the normal vector can be reduced by the equation (2).
In the subsequent step S3, projection processing is performed.
As shown in a schematic diagram of the projection processing in FIG. 4, a projection plane Sp that is perpendicular to the optical axis of the three-dimensional measuring unit 4 and is a virtual two-dimensional coordinate centered on the viewpoint of the three-dimensional measuring unit 4 is used. Then, a projection process is performed.

In detail, it demonstrates below based on the flowchart which shows the projection processing routine shown in FIG.
In step S31, one measurement point is selected from all measurement points. The measurement point selected here is selected from the measurement points not yet selected.

  In step S32, the sphere 2 that detects the measurement point moves in the direction opposite to the direction of the normal vector of the selected measurement point (measurement point moving step). Here, each measurement point constituting the spherical surface of the sphere is almost concentrated at one point which becomes the center of the sphere after the movement, whereas the measurement points of the noise and the non-spherical portion only move by the length of the radius. .

In step S33, it is determined whether all measurement points in the virtual coordinate space have been selected. If the determination result is true (Yes), the process proceeds to step S34. If the determination result is false (No), the process returns to step S31.
In step S34, the moved measurement point group is projected onto a projection plane perpendicular to the optical axis of the three-dimensional measurement means 4 (projection step).

Step S31~S34, for example, as shown in FIG. 6 (A), the measurement point _{P 1} in the direction opposite to the direction of the normal vector _{n P1} of the measurement points _{P 1,} the distance worth measuring point radius r of the sphere P ₁ is moved to be a measurement point A ₁ after the movement. Then, after all the measurement points are moved, the moved measurement point group is projected onto the projection plane Sp as shown in FIG.

Returning to FIG. 2, the sphere center is estimated in step S4.
In detail, it demonstrates below based on the flowchart which shows the estimation routine of the spherical center shown in FIG.
In step S41, the projection plane is partitioned in a grid pattern. Here, the areas are divided so that the areas of the projection planes after the division are substantially equal.
In step S42, the number of super-points in one lattice area on the projection plane divided in step S41 is counted.

Steps S41 and S42 are processes for dividing the projection plane Sp in a grid pattern as shown in FIG. 8A, for example, and counting the number of measurement points after movement included in each grid area Sg.
In step S43, it is determined whether or not all the lattice areas defined in step S41 have been selected. If the determination result is true (Yes), the process proceeds to step S44. If the determination result is false (No), the process returns to step S42.

  In step S44, a grid area (dense area) having a fixed number of measurement points or more after movement is searched from each grid area (dense area search step). As described above, the measurement points after the movement of each measurement point constituting the spherical surface of the sphere are almost concentrated at one point that is the center of the sphere, so that the number of measurement points after movement included in the lattice area is a certain value or more. The region becomes a lattice region including the center of the sphere.

In step S45, as shown in FIG. 8B, measurement after movement in a three-dimensional virtual coordinate space before projection corresponding to the measurement point group after movement included in the lattice area searched in step S44. The center of the sphere is estimated from the point group (sphere center estimation step).
In this step, when the set of measurement points after movement included in the lattice area is S _i , the number of measurement points after movement included in the lattice area is m _i , and the coordinate value of the measurement point after movement is x _i. , the center c _i of the sphere 2 is determined from equation 3 below.

That is, the center c _{i of the} sphere 2 is the average value of the coordinate values x _i of the measurement points after movement.
When the center of the sphere is estimated, the sphere in the virtual coordinate space is detected from the sphere center and radius information, and the detected sphere information is output via the external output means 10.

  Thus, according to the sphere detection method according to the present invention, the normal vector of each measurement point is calculated from the shape data of the sphere 2 acquired as a measurement point group by the three-dimensional measurement means 4, and each measurement point is calculated. Is moved by the radius of the sphere 2 in the direction opposite to the direction of the normal vector, the moved measurement points are projected onto the projection plane, the lattice points where the measurement points after the movement are greater than a certain value are searched for, The coordinates of the measurement point group after movement in the three-dimensional virtual coordinate space before projection corresponding to the measurement point group after movement included in the region are averaged to obtain the center of the sphere to detect the sphere.

Thereby, since the calculation amount performed by the detection of the sphere is proportional to the number of measurement points acquired by the three-dimensional measuring means 4, the calculation amount can be reduced and the sphere can be detected at a higher speed.
In addition, by dividing the projection plane in a lattice shape, the number of measurement points after movement included in each of the partitioned lattice areas is counted, so that a lattice region where the measurement points after movement are dense can be easily searched. Therefore, the sphere can be detected at a higher speed.

Next, modifications of the above embodiment will be described below.
In this modification, the sphere center estimation in step S45 in the sphere center estimation routine shown in FIG. 7 is performed by the method of least squares.

Specifically, after one movement from the measurement point group after movement in the three-dimensional virtual coordinate space before projection corresponding to the measurement point group after movement included in the lattice area searched in step S44 described above. A measurement point is selected, and the center of the sphere in the virtual coordinate space is estimated by performing the least square method on the measurement point after the movement (high accuracy step).
In the case of the above-described modification, by calculating the center of the sphere by the method of least squares, the center of the sphere can be estimated more accurately from the measurement point group after movement, so the detection accuracy of the sphere is further improved. Can be made.

Although the description of the embodiment is finished as described above, the present invention is not limited to the above-described embodiment.
For example, in the above-described embodiment, the projection plane is partitioned in a grid pattern, and the number of measurement points after movement included in each grid area is counted. However, the shape of the projection plane area is not limited to the grid pattern.

  In the above embodiment, the radius information of the sphere 2 is stored in the memory device 8 in advance. However, even when the radius of the sphere 2 is not stored in the memory device 8, the sphere center estimation routine may be performed. When the radius information is not stored in the memory device 8, the radius is calculated from the assumed maximum sphere size, and the predetermined width change determined by the estimation accuracy, the required sphere detection accuracy, etc. The grid area with the maximum number of measurement points after movement is extracted, and the grid area with the largest number of measurement points after movement is extracted from each extracted maximum grid area to estimate the center of the sphere. May be detected.

DESCRIPTION OF SYMBOLS 1 Object recognition apparatus 2 Sphere 4 Three-dimensional measurement means 6 Calculation means 8 Memory device

Claims (4)

A method for detecting a sphere that represents a shape of a sphere as a target in a virtual coordinate space,
A shape data acquisition step of acquiring spherical shape data as a measurement point group by a three-dimensional measurement means;
A normal vector calculation step of calculating a normal vector of each measurement point of the measurement point group constituting the shape data in a virtual coordinate space;
A measurement point moving step of moving each measurement point by the radius of the sphere in a direction opposite to the normal vector;
A projecting step of projecting the moved measurement point moved by the measurement point moving step onto a projection plane perpendicular to the optical axis of the three-dimensional measuring means;
A dense area search step for searching a dense area where the measurement points after movement projected by the projecting step are dense from the projection plane;
A spherical center estimation step for estimating the center of the sphere in the virtual coordinate space from the measurement point group after the movement in the virtual coordinate space before projection corresponding to the dense region searched from the dense region search step;
A method for detecting a sphere, comprising:   The method for detecting a sphere according to claim 1, wherein the dense area search step generates a projection area in which the projection plane is partitioned in a lattice pattern, and searches the dense area from each of the projection areas. .   The method of detecting a sphere according to claim 1, wherein the spherical center estimation step performs estimation by obtaining an average of coordinate values of the measurement point group after movement included in the dense area. .   The sphere center estimation step includes a high-accuracy step of performing estimation by a least square method on the measurement point group after movement included in the dense area of a virtual coordinate space. 3. The method for detecting a sphere according to 2.

Images